frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Qwen 3.8 27B is excellent, but it defaults to overthinking things

https://simonwillison.net/2026/Aug/16/qwen-38-27b/
39•bilsbie•2h ago

Comments

andy99•1h ago
The big problem with overthinking on a dense model is obviously the speed hit you take. Going from Qwen 35BA3B to 27B for me is about 7-8x slower (should be ~9x?). This makes me a lot less patient for useless thinking tokens.

I’d want to compare this to the new Muse 30B model which is super terse and has a whole different way of thinking (no “Wait,”) and in my experiments was way more token efficient to the point that the absolute tok / s didn’t really matter.

simonw•20m ago
Comparing with Muse Glimmer is a good idea. I ran the same exact HTML tool generating prompt against both Glimmer 30B and Qwen 3.8 27B. Results:

Qwen: https://gist.github.com/simonw/121ad098860028b2fab603fa12da1... - 17,576 reasoning tokens, produced this HTML result: https://static.simonwillison.net/static/2026/qwen-over-think...

Glimmer: https://gist.github.com/simonw/51e8ddb2ee597a5005fa63bd4927d... 1,021 reasoning tokens, this HTML: https://static.simonwillison.net/static/2026/glimmer-bbox.ht... - ugly but functional.

In both cases paste in the URL https://static.simonwillison.net/static/2026/two-pelicans-on... to see them work.

Both applications work correctly and fulfill the requirements. The Qwen one (which used the default xhigh reasoning setting) is massively over-engineered. The Glimmer one used whatever their default in LM Studio is and I would argue is a tiny bit under-engineered.

Weirdly the Glimmer one doesn't work with images on other domains like https://static.inaturalist.org/photos/714731804/large.jpg - it fails with a CORS error, but you don't need CORS to load images and detect their width and height, and the Qwen one handles that URL just fine.

That's because Glimmer added this unnecessary line:

  img.crossOrigin = 'anonymous';
bogzz•14m ago
I love reading Glimmer's "thoughts". Why use many word when few do trick?
lostmsu•5m ago
[delayed]
deadcatfound•59m ago
For agents, token efficiency is an operating cost. I’d rather have a terse model that escalates hard cases than one that overthinks every tool call.
SwellJoe•27m ago
This is true, but I think it understates the problem. I did a task I've done with a bunch of small models lately (https://github.com/swelljoe/flar/pull/17), and it did an excellent job, the best of any self-hostable model. But, it took eleven (11!) hours on my dual GPU setup. It really chewed on it, and spent a lot of time checking and re-checking. It is by far the slowest model I've used for the task. GPT 5.5 did a similar task in about 20 minutes. Most big models took about an hour or so, and most small models needed a couple of hours (but did a worse job).
simonw•18m ago
Was that with the default xhigh reasoning setting? I suggest trying again with reasoning set to low or turned off entirely.
SwellJoe•1m ago
Yes, default everything, no tuning, 8_K_XL Unsloth quantization on dual Radeon V620 GPUs (which aren't blazing, but faster than the Strix Halo).
javchz•17m ago
I wonder if this can be fixed with LORAs.
bitexploder•9m ago
I had to fix this on 35B A3B -- I have a proxy that just shuts it down if it gets to 2K thinking tokens and injects something like "We have thought enough, let's begin working." and it almost always finishes the turn then. It rarely needs more than 2K thinking tokens and if it does there is always next turn. I would need to see what 27B is actually doing, but these smaller Qwen models seem prone to this.
cyanydeez•15m ago
--thinking-budget and --thinking-message is all you need in llamacpp to keep it progressing.

the message can be some combination of tool calling, summarizing, etc. It's overthinking often is a bunch of recursion, so simply stopping t and redirecting is all you need to do.

If someones building a harness for llamacpp, you can set this per message, so it's possible to dynamically control it by watching for the expansion of the thinking traces, and redirecting it.

I use the message to tell it to use subagents, add additional logging and to use opencode's dynamic context pruning.

As such, we'll just whisper here _skill issue_.

dofm•8m ago
Unfortunately in xhigh thinking it goes down rabbit holes in such an extreme depth-first way, that whenever you choose to cut it off, there is a very good chance it will not have got round to musing on even half of the prompt! It doesn’t really obviously loop in xhigh, so I am not sure if an “overthinking guard” proxy would have much to go on, but it does obsessively ruminate on edge cases. I have seen it overcomplicate simple code as a result even in my limited testing.

Probably the better solution if you want it to be quicker but still fairly thorough appears to be to configure reasoning effort instead of thinking budget. It seems to do very well still even on the Low setting; on the Medium setting it can get stuck in loops like 3.6 does.

I think xhigh reasoning effort was an absurd choice for a default, and so was not sorting out the chat template so LM Studio could offer the reasoning effort dropdown.

bitexploder•8m ago
Yeah, but be fair. Working with small models is a different ball game. Not all the batteries come included :)
mmastrac•7m ago
I have a private benchmark for disassembly of 80s CPU code and Qwen either does really well or spirals into insanity (looping, failing to call tools). Nemotron is beating it pretty handily, despite being considered a weaker model.

I think it was overtrained and I am starting to suspect that 27B is just not enough to be psychologically stable.

LoganDark•7m ago
I hope Apple does end up moving to HBM. Unified memory has been a huge godsend, but the low memory bandwidth is just such a killer. Even/especially on M5, where the available compute is starting to starve incredibly badly on ML workloads.

The Life and Death of Direct File [pdf]

https://www.ischool.berkeley.edu/sites/default/files/vinton_report_5.pdf
45•ronbenton•1h ago•16 comments

Qwen 3.8 27B is excellent, but it defaults to overthinking things

https://simonwillison.net/2026/Aug/16/qwen-38-27b/
40•bilsbie•2h ago•15 comments

A 3rd World Embedded Engineer Responds to "RISC-V They Should Have Known Better"

https://rvembedded.com/blog_post/12/
376•Narishma•8h ago•198 comments

Rhombus 1.1 is now available

https://blog.racket-lang.org/2026/08/rhombus-v1.1.html
11•spdegabrielle•1h ago•5 comments

Claude: System Prompts

https://platform.claude.com/docs/en/release-notes/system-prompts
542•tosh•13h ago•227 comments

SIMD in the 90s: Programming Intel's Pentium MMX

https://pikuma.com/blog/programming-intel-pentium-mmx-simd
73•ibobev•3d ago•32 comments

Low-Tech Ceramic Water Filter

https://wiki.lowtechlab.org/wiki/Filtre_%C3%A0_eau_c%C3%A9ramique/en
92•Bluestein•5d ago•25 comments

Interview with Amit Patel, Creator of "Solar Realms Elite"

https://breakintochat.com/blog/2013/02/18/amit-patel-creator-of-solar-realms-elite/
9•bananaboy•1w ago•1 comments

Reticulum – Decentralized Mesh Network

https://reticulum.network/
11•sudo_cowsay•1h ago•1 comments

Models Are Getting Dumber on Purpose

https://w4g1.dev/blog/models-are-getting-dumber-on-purpose
270•hruvhwe•6h ago•152 comments

The AI Credit Resale Economy

https://vectoral.com/blog/who-are-the-token-brokers
225•mlenhard•11h ago•89 comments

A quick look at zero-knowledge proofs

https://bernsteinbear.com/blog/zkp/
46•evakhoury•2d ago•18 comments

The federal keyword lists that canceled billions in research funding

https://www.highereddive.com/news/inside-the-federal-keyword-lists-that-canceled-billions-in-rese...
44•walrus01•1h ago•2 comments

Protobuf has LSP support. You're welcome

https://buf.build/blog/protobuf-lsp
109•theanonymousone•7h ago•80 comments

MathCode, Mathematical Coding Agent

https://math-ai-org.github.io/mathcode/
61•homarp•7h ago•19 comments

Dancing with friends and enemies: boids' swarm intelligence

https://community.wolfram.com/groups/-/m/t/122095
10•surprisetalk•5d ago•1 comments

Clamiga: Common Lisp for the Amiga

https://nnamgreb.de/blog/Clamiga+-+Common+Lisp+for+the+Amiga
85•emptybits•3d ago•10 comments

Nvidia dramatically reduces amount of OpenAI infra financing it may guarantee

https://www.reuters.com/business/nvidia-scales-back-250-billion-openai-data-center-guarantee-wsj-...
102•root-parent•4h ago•29 comments

Anton Chekhov played at love most of his life

https://commonreader.wustl.edu/winning-and-losing-at-the-great-game-of-intimacy/
63•lermontov•2d ago•13 comments

Plastic mechanical computer from 1963: The Digi-Comp 1 [video]

https://www.youtube.com/watch?v=-y8bGBE71yw
48•tobr•1d ago•14 comments

Firefox for iOS now has a native adblocker

https://support.mozilla.org/en-US/kb/block-ads-firefox-ios
538•pentagrama•12h ago•221 comments

Stripe Clinches over $7B Deal to Buy AI Firm OpenRouter

https://www.bloomberg.com/news/articles/2026-08-16/stripe-nears-deal-to-buy-ai-firm-openrouter-fo...
199•zacharyozer•5h ago•145 comments

Tell HN: Cloudflare silently injects its analytics when you switch nameservers

285•stagas•8h ago•69 comments

Red queen hypothesis – a new way forward for self-improving AI

https://www.cst.cam.ac.uk/news/red-queen-hypothesis-new-way-forward-self-improving-ai
11•hardlianotion•5h ago•1 comments

Tasklet (YC P26) Is Hiring a Head of Design Engineering

https://tasklet.ai/careers/head-of-design-engineering
1•mayop100•12h ago

A True Telnet BBS on a Casio Calculator

https://ei3lh.eu/2026/08/16/a-true-telnet-bbs-on-a-casio-calculator/
86•austinallegro•13h ago•9 comments

St Lucie Nuclear Reactor Unit 1 manually shutdown, 3 control rods drop into core

https://www.wptv.com/news/treasure-coast/region-st-lucie-county/saint-lucie-nuclear-power-plant-u...
158•toomuchtodo•10h ago•124 comments

A SAT Attack on Tarski's High School Algebra Problem

https://arxiv.org/abs/2608.08421
84•matt_d•4d ago•34 comments

Before Rightmove, there was the Cosmorama

https://www.ianvisits.co.uk/articles/before-rightmove-there-was-the-cosmorama-londons-forgotten-p...
25•brod_ie•5d ago•3 comments

Chestnut – eGPU dock with open-source firmware

https://hwbusters.com/news/comma-ai-egpu-dock-runs-open-source-firmware-249-bare-799-with-an-rx-9...
134•txrx0000•3d ago•39 comments